akm-cli 0.7.0 → 0.7.2

package/dist/tests/github.test.js

@@ -1,161 +0,0 @@
-import { afterEach, describe, expect, mock, spyOn, test } from "bun:test";
-import * as childProcess from "node:child_process";
-import { asRecord, asString, GITHUB_API_BASE, githubHeaders } from "../src/integrations/github";
-// ── Environment helpers ─────────────────────────────────────────────────────
-const originalGithubToken = process.env.GITHUB_TOKEN;
-const originalGhToken = process.env.GH_TOKEN;
-afterEach(() => {
-    mock.restore();
-    if (originalGithubToken === undefined) {
-        delete process.env.GITHUB_TOKEN;
-    }
-    else {
-        process.env.GITHUB_TOKEN = originalGithubToken;
-    }
-    if (originalGhToken === undefined) {
-        delete process.env.GH_TOKEN;
-    }
-    else {
-        process.env.GH_TOKEN = originalGhToken;
-    }
-});
-// ── GITHUB_API_BASE ─────────────────────────────────────────────────────────
-describe("GITHUB_API_BASE", () => {
-    test("is the GitHub API URL", () => {
-        expect(GITHUB_API_BASE).toBe("https://api.github.com");
-    });
-});
-// ── githubHeaders ───────────────────────────────────────────────────────────
-describe("githubHeaders", () => {
-    test("includes Accept and User-Agent headers", () => {
-        delete process.env.GITHUB_TOKEN;
-        const headers = githubHeaders();
-        expect(headers.Accept).toBe("application/vnd.github+json");
-        expect(headers["User-Agent"]).toBe("akm-registry");
-    });
-    test("does not include Authorization when GITHUB_TOKEN is unset", () => {
-        delete process.env.GITHUB_TOKEN;
-        delete process.env.GH_TOKEN;
-        spyOn(childProcess, "spawnSync").mockReturnValue({ status: 1, stdout: "" });
-        const headers = githubHeaders();
-        expect(headers.Authorization).toBeUndefined();
-    });
-    test("includes Authorization when GITHUB_TOKEN is set", () => {
-        process.env.GITHUB_TOKEN = "ghp_test_token_123";
-        const headers = githubHeaders();
-        expect(headers.Authorization).toBe("Bearer ghp_test_token_123");
-    });
-    test("trims whitespace from GITHUB_TOKEN", () => {
-        process.env.GITHUB_TOKEN = "  ghp_trimmed  ";
-        delete process.env.GH_TOKEN;
-        const headers = githubHeaders();
-        expect(headers.Authorization).toBe("Bearer ghp_trimmed");
-    });
-    test("does not include Authorization when GITHUB_TOKEN is empty", () => {
-        process.env.GITHUB_TOKEN = "";
-        delete process.env.GH_TOKEN;
-        const headers = githubHeaders();
-        expect(headers.Authorization).toBeUndefined();
-    });
-    test("does not include Authorization when GITHUB_TOKEN is whitespace-only", () => {
-        process.env.GITHUB_TOKEN = "   ";
-        delete process.env.GH_TOKEN;
-        const headers = githubHeaders();
-        expect(headers.Authorization).toBeUndefined();
-    });
-    test("uses GH_TOKEN when GITHUB_TOKEN is unset", () => {
-        delete process.env.GITHUB_TOKEN;
-        process.env.GH_TOKEN = "ghs_from_gh_token";
-        const headers = githubHeaders();
-        expect(headers.Authorization).toBe("Bearer ghs_from_gh_token");
-    });
-    test("prefers GITHUB_TOKEN over GH_TOKEN", () => {
-        process.env.GITHUB_TOKEN = "ghp_preferred";
-        process.env.GH_TOKEN = "ghs_fallback";
-        const headers = githubHeaders();
-        expect(headers.Authorization).toBe("Bearer ghp_preferred");
-    });
-    test("falls back to gh auth token when env vars are unset", () => {
-        delete process.env.GITHUB_TOKEN;
-        delete process.env.GH_TOKEN;
-        const spawnSyncSpy = spyOn(childProcess, "spawnSync").mockReturnValue({
-            status: 0,
-            stdout: "gho_cli_token\n",
-        });
-        const headers = githubHeaders();
-        expect(headers.Authorization).toBe("Bearer gho_cli_token");
-        expect(spawnSyncSpy).toHaveBeenCalledWith("gh", ["auth", "token"], {
-            encoding: "utf8",
-            timeout: 5000,
-            stdio: ["ignore", "pipe", "ignore"],
-        });
-    });
-    test("does not include gh auth token for non-GitHub URLs", () => {
-        delete process.env.GITHUB_TOKEN;
-        delete process.env.GH_TOKEN;
-        spyOn(childProcess, "spawnSync").mockReturnValue({ status: 0, stdout: "gho_cli_token\n" });
-        const headers = githubHeaders("https://example.com/file.tgz");
-        expect(headers.Authorization).toBeUndefined();
-    });
-});
-// ── asRecord ────────────────────────────────────────────────────────────────
-describe("asRecord", () => {
-    test("returns object as-is for a plain object", () => {
-        const obj = { key: "value", num: 42 };
-        expect(asRecord(obj)).toBe(obj);
-    });
-    test("returns empty object for null", () => {
-        expect(asRecord(null)).toEqual({});
-    });
-    test("returns empty object for undefined", () => {
-        expect(asRecord(undefined)).toEqual({});
-    });
-    test("returns empty object for a string", () => {
-        expect(asRecord("hello")).toEqual({});
-    });
-    test("returns empty object for a number", () => {
-        expect(asRecord(42)).toEqual({});
-    });
-    test("returns empty object for a boolean", () => {
-        expect(asRecord(true)).toEqual({});
-    });
-    test("returns empty object for an array", () => {
-        expect(asRecord([1, 2, 3])).toEqual({});
-    });
-    test("returns the object for nested objects", () => {
-        const nested = { a: { b: "c" } };
-        const result = asRecord(nested);
-        expect(result).toBe(nested);
-        expect(result.a).toEqual({ b: "c" });
-    });
-});
-// ── asString ────────────────────────────────────────────────────────────────
-describe("asString", () => {
-    test("returns string for a non-empty string", () => {
-        expect(asString("hello")).toBe("hello");
-    });
-    test("returns undefined for an empty string", () => {
-        expect(asString("")).toBeUndefined();
-    });
-    test("returns undefined for null", () => {
-        expect(asString(null)).toBeUndefined();
-    });
-    test("returns undefined for undefined", () => {
-        expect(asString(undefined)).toBeUndefined();
-    });
-    test("returns undefined for a number", () => {
-        expect(asString(42)).toBeUndefined();
-    });
-    test("returns undefined for a boolean", () => {
-        expect(asString(true)).toBeUndefined();
-    });
-    test("returns undefined for an object", () => {
-        expect(asString({ toString: () => "obj" })).toBeUndefined();
-    });
-    test("returns undefined for an array", () => {
-        expect(asString(["hello"])).toBeUndefined();
-    });
-    test("returns string with whitespace preserved", () => {
-        expect(asString("  spaced  ")).toBe("  spaced  ");
-    });
-});

package/dist/tests/graph-boost-ranking.test.js

@@ -1,305 +0,0 @@
-/**
- * Integration test for the search-time graph boost (#207).
- *
- * Asserts deterministically that for a corpus with a known graph-eligible
- * query, the rank of the expected target improves when a `graph.json`
- * artifact is present versus absent. No LLM calls are made — the graph
- * file is written directly to the fixture stash, simulating what the
- * extraction pass would produce.
- *
- * The test uses TWO independent runs against the SAME database state:
- *   1. Baseline — `graph.json` deleted before search.
- *   2. Boosted — `graph.json` present before search.
- *
- * Acceptance: the target's rank in the boosted run must be ≤ its rank in
- * the baseline run, and at least one of (rank improves) OR (score
- * strictly increases) must hold. This is a deterministic comparison, not
- * a percentage threshold.
- *
- * It also verifies that the graph signal feeds the SAME `score` field on
- * `SourceSearchHit` — i.e. there is no second SearchHit scorer; the
- * graph-aware run produces a (weakly) higher score on the same hit.
- */
-import { afterAll, beforeAll, describe, expect, test } from "bun:test";
-import fs from "node:fs";
-import os from "node:os";
-import path from "node:path";
-import { akmSearch } from "../src/commands/search";
-import { resetConfigCache, saveConfig } from "../src/core/config";
-import { getDbPath } from "../src/core/paths";
-import { closeDatabase, openDatabase, rebuildFts, setMeta, upsertEntry } from "../src/indexer/db";
-import { GRAPH_FILE_SCHEMA_VERSION, getGraphFilePath } from "../src/indexer/graph-extraction";
-import { buildSearchText } from "../src/indexer/search-fields";
-// ── Environment isolation ───────────────────────────────────────────────────
-let stashDir = "";
-let originalXdgCacheHome;
-let originalXdgConfigHome;
-let originalAkmStashDir;
-let testCacheDir = "";
-let testConfigDir = "";
-beforeAll(() => {
-    originalXdgCacheHome = process.env.XDG_CACHE_HOME;
-    originalXdgConfigHome = process.env.XDG_CONFIG_HOME;
-    originalAkmStashDir = process.env.AKM_STASH_DIR;
-    testCacheDir = fs.mkdtempSync(path.join(os.tmpdir(), "akm-graph-rank-cache-"));
-    testConfigDir = fs.mkdtempSync(path.join(os.tmpdir(), "akm-graph-rank-config-"));
-    stashDir = fs.mkdtempSync(path.join(os.tmpdir(), "akm-graph-rank-stash-"));
-    process.env.XDG_CACHE_HOME = testCacheDir;
-    process.env.XDG_CONFIG_HOME = testConfigDir;
-    process.env.AKM_STASH_DIR = stashDir;
-    resetConfigCache();
-    saveConfig({
-        semanticSearchMode: "off",
-        sources: [{ type: "filesystem", path: stashDir }],
-        registries: [],
-    });
-    buildFixture();
-});
-afterAll(() => {
-    if (originalXdgCacheHome === undefined)
-        delete process.env.XDG_CACHE_HOME;
-    else
-        process.env.XDG_CACHE_HOME = originalXdgCacheHome;
-    if (originalXdgConfigHome === undefined)
-        delete process.env.XDG_CONFIG_HOME;
-    else
-        process.env.XDG_CONFIG_HOME = originalXdgConfigHome;
-    if (originalAkmStashDir === undefined)
-        delete process.env.AKM_STASH_DIR;
-    else
-        process.env.AKM_STASH_DIR = originalAkmStashDir;
-    resetConfigCache();
-    for (const dir of [testCacheDir, testConfigDir, stashDir]) {
-        if (dir)
-            fs.rmSync(dir, { recursive: true, force: true });
-    }
-});
-// ── Fixture builder ─────────────────────────────────────────────────────────
-//
-// The corpus is small but deliberately constructed so the lexical signal
-// is roughly even between two candidates — the graph boost is what
-// separates them deterministically:
-//
-//   - knowledge:database-runbook — recovery procedure for a database
-//     outage. The TARGET of the test query. Doesn't have any postgres-
-//     specific name match, so a postgres-aware query splits between this
-//     and the FAQ on lexical signals alone.
-//   - knowledge:database-faq — Q&A document. Lexical match on the same
-//     query terms but unrelated to outage recovery operationally. The
-//     COMPETITOR.
-//   - memory:incident-2024-shard — operational note. Anchors the graph
-//     edge connecting "outage recovery" to the runbook file.
-//
-// With graph.json present, the runbook's entities directly match the
-// query tokens and pick up the graph boost; the FAQ has no graph node
-// and gets nothing. The deterministic acceptance is "rank improves OR
-// score strictly increases" — both work even if the baseline already
-// happens to put the runbook on top.
-function buildFixture() {
-    // Asset files on disk — the search-time graph boost matches by absolute
-    // file path, so paths must be consistent between fixture build and
-    // graph.json contents.
-    const knowledgeDir = path.join(stashDir, "knowledge");
-    const memoryDir = path.join(stashDir, "memories");
-    fs.mkdirSync(knowledgeDir, { recursive: true });
-    fs.mkdirSync(memoryDir, { recursive: true });
-    const runbookPath = path.join(knowledgeDir, "database-runbook.md");
-    fs.writeFileSync(runbookPath, "---\ntype: knowledge\n---\n\nThe runbook for database outage recovery after a hardware fault.\n");
-    const faqPath = path.join(knowledgeDir, "database-faq.md");
-    fs.writeFileSync(faqPath, "---\ntype: knowledge\n---\n\nA database FAQ covering connection limits, recovery tunables, and outage post-mortems.\n");
-    const memoryPath = path.join(memoryDir, "incident-2024-shard.md");
-    fs.writeFileSync(memoryPath, "---\ntype: memory\n---\n\nDuring the 2024 database outage we recovered shard-3 by following the runbook.\n");
-    // Index the corpus directly into the SQLite DB.
-    const dbPath = getDbPath();
-    // Make sure the cache dir exists (akm-graph-rank-cache-* is fresh).
-    fs.mkdirSync(path.dirname(dbPath), { recursive: true });
-    const db = openDatabase(dbPath);
-    try {
-        const entries = [
-            {
-                entry: {
-                    name: "database-runbook",
-                    type: "knowledge",
-                    filename: "database-runbook.md",
-                    description: "Runbook for database outage recovery after a hardware fault.",
-                },
-                filePath: runbookPath,
-                dirPath: knowledgeDir,
-            },
-            {
-                entry: {
-                    name: "database-faq",
-                    type: "knowledge",
-                    filename: "database-faq.md",
-                    description: "Database FAQ covering connection limits, recovery tunables, and outage post-mortems.",
-                },
-                filePath: faqPath,
-                dirPath: knowledgeDir,
-            },
-            {
-                entry: {
-                    name: "incident-2024-shard",
-                    type: "memory",
-                    filename: "incident-2024-shard.md",
-                    description: "We recovered shard-3 during the 2024 database outage by following the runbook.",
-                },
-                filePath: memoryPath,
-                dirPath: memoryDir,
-            },
-        ];
-        for (const e of entries) {
-            const entryKey = `${stashDir}:${e.entry.type}:${e.entry.name}`;
-            const searchText = buildSearchText(e.entry);
-            upsertEntry(db, entryKey, e.dirPath, e.filePath, stashDir, e.entry, searchText);
-        }
-        rebuildFts(db);
-        setMeta(db, "stashDir", stashDir);
-        setMeta(db, "builtAt", new Date().toISOString());
-        setMeta(db, "stashDirs", JSON.stringify([stashDir]));
-        setMeta(db, "hasEmbeddings", "0");
-    }
-    finally {
-        closeDatabase(db);
-    }
-    // Pre-build the graph file once, but DON'T install it yet — each test
-    // installs/removes it as needed. The entity set is exactly what the
-    // graph-extraction LLM helper would have produced for these bodies.
-    const graphFile = {
-        schemaVersion: GRAPH_FILE_SCHEMA_VERSION,
-        generatedAt: new Date().toISOString(),
-        stashRoot: stashDir,
-        files: [
-            {
-                path: runbookPath,
-                type: "knowledge",
-                entities: ["database", "outage", "recovery", "runbook"],
-                relations: [
-                    { from: "outage", to: "recovery" },
-                    { from: "recovery", to: "database" },
-                ],
-            },
-            {
-                path: memoryPath,
-                type: "memory",
-                entities: ["database", "outage", "recovery", "shard-3"],
-                relations: [{ from: "outage", to: "recovery" }],
-            },
-            // database-faq has NO graph entries — the FAQ doesn't operationally
-            // relate to outage recovery; it just shares vocabulary. This is the
-            // asymmetry that lets the graph signal differentiate the runbook
-            // from a vocabulary-matching FAQ.
-        ],
-    };
-    // Stash the prepared graph payload as a JSON file alongside the stash so
-    // tests can install/uninstall by file rename.
-    fs.mkdirSync(path.join(stashDir, ".akm"), { recursive: true });
-    fs.writeFileSync(path.join(stashDir, ".akm", "graph.prepared.json"), `${JSON.stringify(graphFile, null, 2)}\n`, "utf8");
-}
-function installGraph() {
-    const prepared = path.join(stashDir, ".akm", "graph.prepared.json");
-    fs.copyFileSync(prepared, getGraphFilePath(stashDir));
-}
-function uninstallGraph() {
-    const target = getGraphFilePath(stashDir);
-    if (fs.existsSync(target))
-        fs.unlinkSync(target);
-}
-async function searchHits(query) {
-    const result = await akmSearch({ query, source: "stash", limit: 20 });
-    return result.hits;
-}
-function rankOf(hits, name) {
-    const idx = hits.findIndex((h) => h.name === name);
-    return idx === -1 ? Infinity : idx + 1;
-}
-function scoreOf(hits, name) {
-    const hit = hits.find((h) => h.name === name);
-    return hit?.score ?? 0;
-}
-// ── Tests ───────────────────────────────────────────────────────────────────
-describe("graph boost — search-time integration (#207)", () => {
-    test("graph signal lifts runbook above FAQ for outage-recovery query", async () => {
-        const query = "database outage recovery";
-        // Baseline: no graph file.
-        uninstallGraph();
-        const baselineHits = await searchHits(query);
-        const baselineRank = rankOf(baselineHits, "database-runbook");
-        const baselineScore = scoreOf(baselineHits, "database-runbook");
-        expect(baselineRank).not.toBe(Infinity);
-        // Boosted: graph file present.
-        installGraph();
-        const boostedHits = await searchHits(query);
-        const boostedRank = rankOf(boostedHits, "database-runbook");
-        const boostedScore = scoreOf(boostedHits, "database-runbook");
-        expect(boostedRank).not.toBe(Infinity);
-        // Acceptance criterion (deterministic, not a percentage threshold):
-        //   - rank must not regress, AND
-        //   - score must not regress.
-        // Per CLAUDE.md / spec §9, displayed scores are clamped to [0,1]; on
-        // the strong-match runbook fixture both runs may clamp to the ceiling,
-        // so the observable contract collapses from "score strictly increases"
-        // to "score does not regress" while rank ordering separately confirms
-        // the graph signal lifted the runbook over the FAQ.
-        expect(boostedRank).toBeLessThanOrEqual(baselineRank);
-        expect(boostedScore).toBeGreaterThanOrEqual(baselineScore);
-    });
-    test("absent graph file has no effect on score (no parallel scoring track)", async () => {
-        // With the graph file uninstalled, a non-graph-eligible query should
-        // produce the same hit ordering as it did before #207 landed. This
-        // confirms the graph integration is purely additive — there's no
-        // hidden second scorer running unconditionally.
-        uninstallGraph();
-        const without = await searchHits("database");
-        expect(without.length).toBeGreaterThan(0);
-        // Re-run the same query while the file is uninstalled — must be
-        // byte-identical (same hits, same scores) within the deterministic
-        // tiebreaker.
-        const without2 = await searchHits("database");
-        expect(without2.map((h) => h.name)).toEqual(without.map((h) => h.name));
-        expect(without2.map((h) => h.score)).toEqual(without.map((h) => h.score));
-    });
-    test("score is clamped to [0,1] even when boosts would push above 1.0", async () => {
-        // Fixes a pre-existing breach of the CLAUDE.md / spec §9 contract that
-        // locks `SearchHit.score` to [0,1]: the boost loop in db-search.ts can
-        // accumulate FTS-base + multiple additive boosts whose product exceeds
-        // 1.0, and the addition of #207's graph boost (up to ~1.05 additive)
-        // makes the breach detectable in practice. The runbook fixture above
-        // matches an exact-name query (boost +2.0) AND has a full graph hit,
-        // which combined push the raw computed score above 1.0. The clamp
-        // (`Math.min(1, Math.max(0, score))` near `Math.round(score * 10000)`
-        // in `searchDatabase`) guarantees the final SearchHit.score is exactly
-        // 1 in that case rather than overflowing.
-        installGraph();
-        const hits = await searchHits("database-runbook");
-        const target = hits.find((h) => h.name === "database-runbook");
-        expect(target).toBeDefined();
-        expect(typeof target?.score).toBe("number");
-        // Every emitted score must satisfy the locked contract.
-        for (const h of hits) {
-            expect(h.score ?? 0).toBeLessThanOrEqual(1);
-            expect(h.score ?? 0).toBeGreaterThanOrEqual(0);
-        }
-        // The exact-name + graph-boosted case clamps to the ceiling.
-        expect(target?.score).toBe(1);
-    });
-    test("graph signal feeds the same SearchHit.score field — no second scorer", async () => {
-        // Verify the boost lands on the same `score` property of the same hit
-        // object. This is the contract: the graph signal is one boost
-        // component inside the FTS5+boosts loop, not a parallel ranking.
-        const query = "database outage recovery";
-        uninstallGraph();
-        const baseline = await searchHits(query);
-        const baselineHit = baseline.find((h) => h.name === "database-runbook");
-        expect(baselineHit).toBeDefined();
-        expect(typeof baselineHit?.score).toBe("number");
-        installGraph();
-        const boosted = await searchHits(query);
-        const boostedHit = boosted.find((h) => h.name === "database-runbook");
-        expect(boostedHit).toBeDefined();
-        expect(typeof boostedHit?.score).toBe("number");
-        // Same hit shape, same score field — just (weakly) higher.
-        expect(boostedHit?.path).toBe(baselineHit?.path);
-        expect(boostedHit?.ref).toBe(baselineHit?.ref);
-        expect(boostedHit?.score ?? 0).toBeGreaterThanOrEqual(baselineHit?.score ?? 0);
-    });
-});